FCR-N REPORT

(1)

REPORT

FCR-N DESIGN OF REQUIREMENTS

VERSION 1 - 4 JULY 2017

AUTHORS

ROBERT ERIKSSON SVENSKA KRAFTNÄT

NIKLAS MODIG SVENSKA KRAFTNÄT

ANDREAS WESTBERG SVENSKA KRAFTNÄT

THIS IS A BACKGROUND DOCUMENT FOR THE NEW TECHNICAL REQUIREMENTS FOR FCR IN THE NORDIC

SYNCHRONOUS SYSTEM.

THE REQUIREMENTS ARE SO FAR DRAFT REQUIREMENTS AND ARE THUS NOT REQUIREMENTS TO BE FULFILLED FOR

CURRENT NORDIC FCR MARKET.

(2)

I NTRODUCTION

1.

Frequency quality is a measure of the power balancing in a synchronous system. The balancing is driven by variations in production and consumption, together with the control response of the reserves and inertia. The frequency quality in the Nordic system has reduced over the last years indicated by increased minutes outside normal band. The normal frequency band is 50±0.1 Hz which should not be exceeded more than 10 000 minutes per year. The frequency containment reserve for normal operation (FCR-N) is to handle the short term stochastic net power variation in production and consumption. Recently, secondary control, automatic frequency restoration reserve (aFRR) has been introduced in order to improve the quality but is not seen as the sole solution of the problem. Revision of the FCR-N in order move towards better quality is a complementary solution. One aim of the frequency containment process project (FCP project) is to develop thorough requirements on the FCR-N ancillary service to ensure good frequency quality.

This document describes steps that have been taken to create the new FCR-N requirements.

B

ACKGROUND OF THE PROJECT

1.1

The FCP project stems from a project series named “Measures to mitigate frequency oscillations with a time period of 40-90 s” (commonly known as the 60 s-project), which in turn consisted of phase 1 and phase 2. The 60 s-projects investigated how FCR-N was implemented practically within the Nordic region. The reason was that from a top-down viewpoint it was thought that the specification for FCR-N was fairly transparent and straight forward. From a bottoms-up viewpoint the specifications were though anything but consistent throughout the Nordic synchronous area resulting in various different implementations. It was thus decided that a new requirement shall also try to harmonize the practical implementation of FCR within the Nordic synchronous area.

The 60 s-project also included physical testing of hydro units that provided FCR. Twelve different hydro power stations were tested with various testing procedures such as frequency step response tests and sine-in-sine-out tests. For linear systems one can inject a sinusoidal signal and measure the output, which also is sinusoidal, but its amplitude and phase may have shifted. The

(5)

sine-in-sine-out tests were performed in open loop by injection of an artificial frequency signal in the governor, i.e. a super imposed sinusoidal frequency signal fed into the governor. From this, transfer functions were estimated and stability in closed loop system was analysed. During these tests much new knowledge was gained in how the FCR requirements were implemented in practice, some were good and some were not. An example of the difference in implementation is shown in Figure 1 where in total 39 different sine-sweep tests using the twelve different hydro power stations are presented in a discrete Nyquist-like graphⁱ. Each dashed curve is a test, at a hydro power plant, excited by a set of sinusoidal signals with different time periods injected into the governor. The response of the Nyquist curve at discrete frequencies is marked with ‘x’ and linear interpolation has been applied in between. Ideally, a curve should not enter the black circle and shall not appear on the left hand side of the point -1 (red dot) by encircle this point. Such response acts de-stabilising, clearly, units that act destabilising could be identified. Figure 2 shows two selected responses illustrating one response that stabilises the system (blue curve) and another that de-stabilises (red curve).

iThe graph is created by assuming that all FCR-N providers have the same dynamic response as the tested unit.

(6)

FIGURE 1.RESULT FROM SINE-SWEEP TESTS PERFORMED DURING THE 60S PHASE 1 AND 2 PROJECTS

The results from previous projects show that there is a clear need to physically test and verify the dynamic performance. The existing requirements state that a unit shall have full activation after 150-180 seconds from a stepwise change in the frequency within the normal operating band of the FCR-N. In Sweden there is an additional requirement that 63 % of the steady state power response shall be activated after 60 seconds. The sine-sweep tests clearly showed that there is some unwanted dynamic performance that can be seen with sine-sweep tests but not with step response tests that have previously been used in the Nordic synchronous area.

FIGURE 2.SELECTED RESPONSES FROM REAL TESTING, BLUE RESPONSE ACTS STABILISING AND RED ACTS DE-STABILISING.

S

COPE OF THE PROJECT

–

FCR

-

N

1.2

This work focuses on the design of the requirements of the FCR-N. The design shall consider the system need but is limited to technical limitations in hydro power units. Reasonable amount of hydro power units have to qualify in order to open up for enough capacity on the market and endorse competitive prices. The project aims not to revise the requirements on the aFRR but rather use the existing implementation to find proper shares between FCR-N and aFRR. The

(7)

project will only set the technical requirements for FCR-N and FCR-D products with constraints given. Impact from voltage dependency in loads, distributed inertia and activated network protections schemes are not considered. The analysis of the design is performed only on a linear single-input-single-output system with no voltage dependency but is verified in non-linear simulations.

G

OALS

1.3

The goals are to come up with requirements on the FCR-N. The requirements shall 1. be functional and testable locally at each FCR provider,

2. specify dynamic response to ensure stability,

3. improve the frequency quality in relation to specified key performance indicators (KPIs) in relation to the system of today. The KPIs are to be specified in the project,

4. specify the dynamic response from net power variation to frequency deviation to meet the KPIs.

C

ONSTRAINTS

1.4

Constraints are given in Table 2.

O

UTLINE

1.5

This document is organised as follows. In Section 2 a theoretical background is given for general control systems on the basis of transfer function with the concepts stability and performance.

Next section provides the model description of an FCR-unit and different per unit scaling. The reference hydro power unit is also introduced which is used throughout the document. In Section 4 the requirements are described and motived, also results from simulations are provided. Section 5 gives an overview of real test procedure on site and how to achieve the response in order to verify qualification.

T HEORETICAL B ACKGROUND

2.

In most physical systems non-linearities are to be considered in the control design. However, linear control design is often used and then verified by non-linear simulations and testing. This

(8)

section describes linear control design for a single-input-single-output (SISO) system and how the non-linearity introduced by backlash can be included.

S

TABILITY

2.1

First some terminology on stability is used in terms of asymptotic stability which means that there exists no initial condition or no bounded input signal that drives the output to infinity.

Figure 3 shows a SISO system where F(s) is the transfer function of the control process, G(s) the plant, d is a disturbance signal entering the system, y is the output of the closed loop system and s is the Laplace operator.

F(s)F(s) G(s)G(s) -

output

∑ ∑

disturbance d

system Control

unit

y

FIGURE 3.OVERVIEW OF A FEEDBACK SYSTEM

The aim is to determine whether or not the closed loop system is stable. The mathematical framework of transfer functions provides an elegant method, which is called loop analysis. The basic idea of loop analysis is to trace how a sinusoidal signal propagates in the feedback loop, this by investigating if the propagated signal grows or decays. One way to analyse stability is by using the Nyquist criterion which in turn uses the loop gain. The loop gain is defined as

𝐺_𝑂(𝑠) = 𝐹(𝑠)𝐺(𝑠).

(2.1) The loop transfer function, also named sensitivity, is defined as

𝑆(𝑠) =_1+𝐺¹

O(𝑠) (2.2)

and describes the propagation of a signal through the loop i.e. how the output amplifies through the loop.

The amplification of a signal is determined by the denominator. Whether the signal grows as it is phase shifted by 180^o (the signal has opposite sign) in the loop determines if the system is stable or not. The point where a signal has this phase shift and its amplitude remains (gain equals to one) corresponds to where to denominator is equal to zero i.e.

𝐺_O(𝑠)|_{𝑠=𝑗𝜔}₀ = −1.

(2.3) At such conditions the signal grows to infinity, thus, the point -1 is of interest together with the loop gain.

The Nyquist curve is the loop gain, that can be plotted in the complex plane, with the Laplace operator s replaced by the complex value jω and ω varying as shown in Figure 4. The system is asymptotically stable if the Nyquist curve does not encircle the point -1. Basically, at the point where the Nyquist curve has a phase shift of 180^o the loop transfer function should be smaller

(9)

than one. This holds true for simple enough systems (loop gains) as one could in reality cross the negative real axis twice to the left of the point -1 and still not encircle this point. Note that this is only valid if the loop gain is at least marginally stable i.e. no poles in the right half plane. For a more detailed description readers are referred to [1] and textbooks in the field of linear control theory. In practice it is not enough that a system is stable. There should also be some margins of stability that would describe how stable the system is and its robustness to perturbations. A stability margin is introduced by a distance between the Nyquist curve and the point -1. It can be specified in terms of amplitude marginⁱⁱ (also known as gain margin), (Am), phase marginⁱⁱⁱ, (ϕm), and the smallest Euclidian distance, r, between the curve and the point -1 (referred to as the stability margin).

FIGURE 4.NYQUIST DIAGRAM.NOTE THAT THE INDICATED PHASE AND GAIN MARGIN ARE HERE IMPOSED BY THE CIRCLE. THE BLUE CURVE HAS LARGER MARGIN THAN GUARANTEED BY THE EUCLIDIAN DISTANCE.

However, specifying the Euclidian norm guarantees that the amplitude and phase become as follows

𝐴_m≥ 1

1 − 𝑟 (2.4)

𝝋_𝐦≥ 𝟐 𝒔𝒊𝒏^−𝟏(^𝒓_𝟐). (2.5)

A drawback with gain and phase margins is the necessity to state both of them in order to guarantee the Nyquist curve not to become close to the critical point. Moreover, phase and amplitude margin, stand-alone or combined, do not guarantee a certain distance to the point -1.

Note that none of the mentioned margins guarantee closed loop stability themselves – the point -1 may be encircled without entering the r-circle, and both the unit circle and the negative real axis may be crossed multiple time. However, it can be assumed that the loop gain is simple enough so that such margins ensure stability.

ii the factor by which the loop gain can be increased until the Nyquist curve intersects with the point -1+ 0∙j

iii angle between the negative real axis and the point where the curve crosses a circle centred in origin with unity radius.

G0(jω)=

F(jω)G(jω)

(10)

The stability margin expressed by the Euclidian norm limits the sensitivity function below a certain value as follows

|𝑆(𝑠)| ≤ ¹_𝑟= 𝑀_s (2.6)

as the sensitivity function is the loop transfer function.

This comes from the fact that the denominator in the sensitivity function is the Euclidian distance between the loop gain and the point -1.

Thus, keep the supremum norm (𝑚𝑎𝑥_∀𝜔|𝑆(𝑗𝜔)|) below one over r ensures the loop gain not to amplify more at any particular frequency. If nominal stability is fulfilled, i.e. the point -1 in the Nyquist plane is not encircled it implies robust stability and implies uncertainties to be allowed in the plant or controller.

R

EJECTION OF DISTURBANCE

2.2

The transfer function from a disturbance entering the system is given by

𝐺(𝑠)

1+𝐺0(𝑠)𝑑 = 𝑆(𝑠)𝐺(𝑠)𝑑 = Δ𝑓 (2.7)

where Δ𝑓 is the output of the close loop system.

Thus, the transfer function from a disturbance is the sensitivity function times the transfer function of the system. Therefore, the sensitivity function not only matters in the stability analysis but also plays an important role in how a disturbance propagates in the system. Moreover, the transfer function is independent on the modelling of the disturbance signal - deterministic or stochastic.

F(s)F(s) G(s)G(s) e

-

output D(s)D(s)

∑ ∑

disturbance

∆f d

system FCR unit

FIGURE 5.OVERVIEW OF THE SYSTEM

Main point: To ensure robust stability it is equivalent to check either the Nyquist curve or the maximum sensitivity (2.6)

(11)

Now consider a disturbance, e(t)=sin(ωt) and |e(jω)|=1, that enters the system through a filter D(s), shown in Figure 5. Then, assume the control object to be |∆f(jω)|<1 for all ω. From (2.7) the following can be derived

|𝑆(𝑗𝜔)| <[𝐷(𝑗𝜔)𝐺(𝑗𝜔)]¹ (2.8)

Note that this describes how a signal that enters the system does not propagate in the system so its amplitude is larger than the initial value of the disturbance for any frequency. However, the system is linear and the output is obtained by superposition of the signals that have propagated through the system. If the disturbance signal contains several frequencies, e.g. stochastic signals, they interrelate and may therefore result in input amplitude larger than one even though the power is very low at particular frequencies. Therefore, since the system is linear the output is obtained by superposition. The bottom-line is, the output may therefore also be larger than one.

Figure 6 shows a random signal generated by a Gaussian white noise process. White noise has a power spectrum of one and the probability of a random sample to occur outside ±1 is about 32%.

FIGURE 6.STOCHASTIC SIGNAL – WHITE NOISE

M ODEL DESCRIPTION

3.

This section provides a description of the power system (G(s)) and the control unit (F(s)) models used. The system consists of generation and consumption distributed in the grid. Thus inertia and

(12)

frequency control is distributed and connected through the grid. In this project, the modelling of the power system and the FCR providing units is performed with the one machine equivalent, assumptions for this are given below.

P

OWER SYSTEM MODEL

3.1

The swing equation relates the rotor dynamics with mechanical and electrical power of a single machine as

𝑯𝒙

𝝅𝒇_𝟎 𝒅^𝟐𝜽𝒙

𝒅𝒕^𝟐 = 𝑷_𝒎𝒙− 𝑷_𝒆𝒙 (3. 1)

where 𝜃_𝑥 is the angle in rad of generator x, 𝐻_𝑥 is the inertia constant, Pmx and Pex are the mechanical and electrical powers, respectively, expressed on a power base. 𝑓₀ is the nominal frequency. Consider the synchronous machines on a common system base (𝑆_n). Assume the machine rotors swing coherently, i.e. all 𝑑𝜃_𝑥/𝑑𝑡 are equal, the powers and the dynamics can then be added as

∑ 𝐻_𝑥 𝜋𝑓₀

𝑑²𝜃_𝑥 𝑑𝑡²

𝑥

= ∑(𝑃_𝑚𝑥

𝑥

− 𝑃_𝑒𝑥) (3.2)

This results in 𝐻

𝜋𝑓₀ 𝑑²𝜃

𝑑𝑡² = 𝑃_𝑚 − 𝑃_𝑒 (3.3)

where the equivalent inertia constant H for the complete system is given by

𝐻 = ∑ 𝐻_𝑥∀𝑥 (3.1)

where 𝐻_𝑥 is the inertia constant of generator x on this common power base.

Loads are here modelled not to depend on voltage; therefore, they can be lumped. The static loads are assumed to be frequency dependent in proportion to the frequency deviation. Thus, the one mass model together with load frequency dependency then relates the transfer function from power change to frequency change as

Δ𝑓 = 𝑓₀

(2𝐸_k𝑠 + 𝑆_n𝑘𝑓₀)∆𝑃 = 𝐺(𝑠)∆𝑃 (3.2)

where parameters are specified in Table 2. Thus, this transfer function is a single-input-single- output (SISO) model of the power system. Voltage dependency is not considered as it requires more detailed modelling than the one mass model to properly capture the dynamics. In addition, the view here is that frequency is not strongly correlated with voltage variations.

(13)

C

ONTROLLED UNIT

3.2

Distributed control implies several controlled units to contribute to the total control work. Figure 7 illustrates several control vectors and the total vector at a particular frequency ω. A unit x provides a response 𝐹_𝑥(𝑗𝜔) and the total sums to

𝐹(𝑗𝜔) = ∑ 𝐹_𝑥 _𝑥(𝑗𝜔). (3.3)

However, here it is assumed that the summed response acts on the coherently swinging system,

explained above. The control response is therefore here assumed to be delivered by a single unit which control response is scaled to correspond to the total regulating strength in the system.

Figure 8 shows several units in parallel providing control response to the system.

FIGURE 8.OVERVIEW OF SEVERAL FCR-N PROVIDERS IN PARALLEL.

𝐹(𝑗𝜔)

𝐹_𝑥(𝑗𝜔)

FIGURE 7:TOTAL CONTROL RESPONSE FROM SEVERAL PROVIDERS

(14)

R

EFERENCE FCR

-

UNIT

–

HYDRO POWER

3.2.1

In this project a simplified hydro power unit is used as a reference unit in order to come up with requirements that also will qualify enough capacity in the Nordic market. The project uses the so called “Nordic frequency model” which is a linear model with backlash on top. An overview of the linear hydro power plant model (F(s)) is displayed in Figure 9. Parameters are defined in Tables 1 and 2. The model is built up of a controller with proportional and integrator part (so called PI- controller), a servo modelled by a low pass-filter in feedback with the droop and the penstock. The penstock provides important and limiting dynamics which are of non-minimum phase. The non- minimum phase dynamic puts limitation on the closed-loop bandwidth but this is not further discussed here.

FIGURE 9.LINEAR AGGREGATED REFERENCE MODEL.

The rating and droop value play a role in the provision of FCR, the scaling of individual units’

response are explained in Subsection 3.3. Note that the droop value in combination of the power base of the unit defines the regulating strength.

Further details of the hydro power plant modelling can be found in the description of the Nordic frequency Model [2].

Backlash has been added in the model shown in Figure 10 which is added after the servo.

FIGURE 10.NON-LINEAR AGGREGATED REFERENCE MODEL.

(15)

The internal feedback can also be by feedback of the power, thus, the backlash is compensated and its impact reduces. The impact from backlash is illustrated in Figure 12 which shows an input signals that passes through a backlash which affects both the phase and amplitude of the output.

Note that we here define backlash as ±b. An input with amplitude A then the output reduces to A- b (if b<A and the backlash centred).

Figure 11 shows the phase shift as function of backlash and is the ratio between the fundamental components of the input and output. Figure 12 provides the phase shift in the time domain and is provided by calculating the fundamental component (through Fast Fourier transform – FFT) of the output signal. Note that only the backlash is considered here, if the input signal passes through an LTI block before it enters the backlash additional phase shift adds up.

FIGURE 11.PHASE SHIFT AS FUNCTION OF THE RATIO BETWEEN THE BACKLASH AND SIGNAL AMPLITUDE.

(16)

FIGURE 12.PHASE SHIFT FOR DIFFERENT BACKLASH VALUES

Applying the Fast Fourier transform (FFT) of a sinusoidal signal with amplitude A results in an amplitude of A at the particular frequency of the sinusoidal. However, the FFT of the output signal with backlash results in an amplitude larger than A-b. Thus, the impact of backlash is indirectly reflected in linear analysis as it is included in the output response. The fundamental component scaling takes into account the fact that the amplitude of the fundamental component is larger than the actual signal, as shown in Figure 13.

(17)

FIGURE 13.INFLUENCE OF BACKLASH ON A SINUSOIDAL SIGNAL.

The calculation of the fundamental scaling factor, α, is done based on the size of the backlash in relation to the signal strength. In a block diagram in Figure 14 the calculation is shown. In the sine- in-sine-out tests the unit is trying to provide a signal 𝑎(𝑡) but due to the backlash the signal will not be purely sinusoidal. The signal 𝑦(𝑡) represents the output signal due to backlash. Signals a(t) and y(t) are the blue and red curves in Figure 13, respectively. 𝑎(𝑡) is given as

𝑎(𝑡) = sin(𝜔₀𝑡)

(3.4) where 𝜔₀is an arbitrary frequency > 0.

Now the output signal 𝑦(𝑡) can be simulated with a(t) entering the backlash block, shown in Figure 14. The discrete FFT is calculated of the input and output signals as

𝐴(𝑘) = ∑ 𝑎(𝑛)

𝑁−1

0

𝑒^{−𝑗2𝜋𝑛𝑘}^𝑁 (3.5)

𝒀(𝒌) = ∑ 𝒚(𝒏)

𝑵−𝟏

𝟎

𝒆^{−𝒋𝟐𝝅𝒏𝒌}^𝑵

(3.6) 𝑇𝑜𝑡𝑎𝑙 𝐵𝑎𝑐𝑘𝑙𝑎𝑠ℎ_𝑝𝑢

𝑎(𝑡) 𝑦(𝑡)

FIGURE 14.BLOCK DIAGRAM OF THE TOTAL BACKLASH

(18)

The fundamental scaling factor can be calculated after performing the Fourier transforms.

𝛼 =𝑌(𝜔₀) 𝐴(𝜔₀)

(3.7)

Table 3 shows different backlash values and the corresponding scaling factor, α.

TABLE 3.BACKLASH SCALING FACTOR (𝜶) AS A FUNCTION OF TOTAL BACKLASH IN PERCENT OF TOTAL SIGNAL STRENGTH (±𝒃)

2𝑏 0 % 1 % 2 % 3 % 4 % 5 % 6 %

α 1 0.999 0.998 0.997 0.996 0.994 0.992

2𝑏 7 % 8 % 9 % 10 % 11 % 12 % 13 %

α 0.99 0.988 0.986 0.984 0.981 0.979 0.976

2𝑏 14 % 15 % 16 % 17 % 18 % 19 % 20 %

α 0.974 0.971 0.968 0.965 0.962 0.959 0.956

2𝑏 21 % 22 % 23 % 24 % 25 % 26 % 27 %

α 0.953 0.95 0.946 0.943 0.94 0.936 0.932

2𝑏 28 % 29 % 30 %

α 0.929 0.925 0.921

The internal feedback, ep, in an FCR-unit is most often expressed in percentage and is called droop.

This percentage value is expressed on its own power base, most often rated power, and is defined as

𝑒_𝑝 =

𝑑𝑓⁄𝑓₀

∆𝑃⁄𝑆n−𝐹𝐶𝑅. (3.8)

This equation states, a unit changes its power by 100 % at frequency change of ep [%]. Example, a unit with droop of 6 % requires the frequency to drop 0.06 ∙ 50 𝐻𝑧 = 3 𝐻𝑧 to change its power by 100 %. Droops are commonly in the range 2-12% which implies the power change for FCR-N (±0.1 Hz) is in the range of ±10 % to ±1.67 % of the machines’ power base.

Backlash reduces the output and a typical backlash value of hydro power plants in the Nordic system is around ±0.005 pu^iv [3]. Note that this value is given on the power base of the machine and is independent of the droop. However, the ratio backlash divided by FCR-N capacity is strongly depended on the droop. In order to achieve the required total steady-state capacity backlash is compensated according to

Δ𝑃_FCR = 𝑆_n−FCR[𝑑𝑓 𝑓₀ ∙ 1

𝑒_p− 𝑏] (3.9)

ivBased on the machine’s power base

(19)

where Δ𝑃_FCRis the FCR capacity.

P

ER UNIT SCALING

3.3

Until now physical units have been used for the system input and output. These physical inputs and outputs can be scaled to per unit (pu). For the FCR-N three different per unit scaling have been used, these are

1. Per unit – droop base

Scaling is based on the droop value and scales the rated power (or number of machines) to deliver 600 MW FCR-N in total – Used in the simulation study.

2. A. Per unit – FCR-N capacity base

Scaling is based on FCR-N delivery, the power and frequency deviation bases are 600 MW and 0.1 Hz, respectively – Used in the control design.

B. Per unit – machine base

Scaling is based on the capacity delivered from an individual unit. The power base comes from the frequency step responses where the static gain is equal to one per unit – Used in the actual testing.

Hence, it is important to know on which base the per unit values refers to as they may seem to be similar. The beauty with per unit is that individual units’ responses are scaled to one whereas the system also scales to one which then makes them compatible without further scaling. More detailed description is given below.

P

ER UNIT SCALING

–

DROOP BASED

3.3.1

This per unit scaling uses given droop values as the base to derive the power base. For this, there are now two options, the power base of a single machine is scaled to deliver the full capacity of FCR-N based on the selected droop or the machine power base is also chosen and the number of machines is scaled to deliver the right amount of FCR-N.

The rated power of a single FCR unit delivering the capacity 𝑑𝑃with a droop of 𝑒_𝑝corresponds to

𝑆_n−FCR = 𝑒_P∙ 𝑑𝑃 ∙_𝑑𝑓^𝑓⁰. (3.10)

The capacity of 600 MW using droop 𝒆_𝒑 =6 % gives the rated power as 𝑆_n−FCR6%= 𝑒_P∙ 𝑑𝑃 ∙_𝑑𝑓^𝑓⁰ = 0.06 ∙ 600 ∙_0.1H^50H^Z

ZMVA = 18000MVA. (3.11)

Alternatively, each unit on the base Sn-FCR, using droop of 6 %, delivers a capacity of

(20)

Δ𝑃 = 𝑆N−FCR∙^𝑑𝑓_𝑓

0 ∙_𝑒¹

P (3.12)

with similar numbers as above each unit delivers Δ𝑃 = 𝑆_N_−FCR∙^𝑑𝑓_𝑓

0 ∙_𝑒¹

P= 600 ∙^0.1₅₀∙_0.06¹ = 20MW. (3.13)

To deliver the full amount, n number of units is required which calculates as

𝑛 =_Δ𝑃^𝑑𝑃. (3.14)

Simulation runs performed in the Nordic frequency model are based on this per unit scaling. Since the model, backlash excluded, is linear one single machine can be used to provide the whole capacity.

A. P

ER UNIT SCALING ON SYSTEM LEVEL

3.3.2

The SISO-model in Figure 15 has physical units for input/output (I/O). By defining the power base as

𝑑𝑃 = 600 𝑀𝑉𝐴 = 1 𝑝𝑢

(3.15) and the frequency deviation base as

𝑑𝑓 = 0.1 𝐻𝑧 = 1 𝑝𝑢.

(3.16)

FIGURE 15.A(SISO) FREQUENCY CONTROL MODEL.SEE TABLE 1 FOR VARIABLE DEFINITION.

The input and output static gain for the FCR-provider becomes one per unit. This step of the scaling can be achieved since the droop 𝑒_𝑝 is defined as

𝑒_p =

d𝑓⁄𝑓₀

𝑑𝑃⁄𝑆_n−FCR (3.17)

(21)

where d𝑓 and 𝑓₀ are the frequency deviation limit and nominal frequency, respectively. 𝑑𝑃 is the static capacity of the unit and 𝑆_N_−FCR the rated power of the unit.

FIGURE 16.SISO-MODEL IN FIGURE 15 SCALED FROM PHYSICAL UNITS TO PU I/O

Then the system scales to 𝐺(𝑠) = 𝑑𝑃 ∙ 𝑓₀

𝑑𝑓 2𝐸_k𝑠 + 𝑘𝑓₀𝑆_𝑛,

(3.18)

The response from an FCR-unit is one per unit with droop of 0.002 (0.2 %) and rated power 600 MVA. This simplification of the modelling is shown in Figure 17.

FIGURE 17.SIMPLIFICATION OF SCALED MODEL IN FIGURE 16.

B. P

ER UNIT SCALING

–

MACHINE BASE

3.3.3

The normalization is defined so that the static gain of an FCR unit shall be equal to one per unit, i.e.

𝐹𝐹𝑇[𝐹(𝑗𝜔)] = 1 pu 𝜔 → 0. (3.19)

Such scaling is performed by incorporating the backlash and the fundamental component of the output signal. Further description is given in Section 5.

(22)

D ESIGN OF REQUIREMENTS

4.

The goals of the design are to improve the frequency quality and to ensure stability. The main measure of frequency quality is minutes outside normal band (MoNB) – mathematically defined as

𝑘𝑝𝑖_{𝑀𝑜𝑁𝐵} =

{∫ 𝑑𝑡, |∆𝑓(𝑡)| > 0.1 𝐻𝑧 0, |∆𝑓(𝑡)| < 0.1 𝐻𝑧

∫ 𝑑𝑡 . (4. 1)

This KPI is mentioned in the constraints and specifies that MoNB should be less than 10 000 minutes per year corresponding to 1.9 % of the time.

The methodology developed within this project, for creating the future requirements imposed on the FCR-N, is based on linear design considering fundamental limitations [4]. Stability and performance are expressed with and without uncertainty as

 Nominal stability: The system is stable with no model/control uncertainty

 Nominal performance: The system satisfies the performance specifications with no model/control uncertainty

 Robust stability: The system is stable for all perturbed plants/controller about the nominal model up to the worst-case model uncertainty.

 Robust performance: The system satisfies the performance specifications for all perturbed plants/controllers about the nominal model up to the worst-case model uncertainty.

The project applies robust stability for the low inertia system in addition to low frequency dependency of loads. This implies an uncertainty for the low inertia system, given by the Euclidian distance, is allowed before instability. Thus, there is an uncertainty margin which can either be in the plant or in the FCR-unit response. This can be realised from the Nyquist curve as the loop gain is defined by the system response times the controller response.

Moreover, the project has chosen to use nominal performance which means performance meets the requirements in the average inertia system without uncertainty and provides acceptable frequency quality on average. However, it is likely that the power disturbance varies over the year but not necessarily correlated with the variation in the inertia of the system.

Hence, stability is expressed on the low inertia system and performance is expressed on the average inertia system. The project believes that by using nominal performance and robust stability a significant policy step forward is made compared to the current situation in the Nordic synchronous area. In the future, an enhancement of this policy would be to move from nominal performance towards robust performance.

(23)

S

TABILITY REQUIREMENT

4.1

The constraints, see Section 1, specify minimum stability requirements in terms of phase margin.

Using this margin the maximum sensitivity Ms is calculated from (2.6) as 𝑀_𝑠 =¹_𝑟 = ¹

2∙sin(^𝜑m₂ ∙₁₈₀^𝜋 )= ¹

2∗sin(²⁵₂∗₁₈₀^𝜋 ) = 2.31. (4.2)

Thus, a circle with radius 1/Ms is plotted in the Nyquist diagram centred at the point -1. For accepted control response, i.e. robust stability, the Nyquist curve shall not enter the circle or encircle the point -1. However, so far the response is on the complete FCR response and not on unit level. This can also be expressed as a requirement on the sensitivity function as

|𝑆_min(𝑠)|_∀𝜔 < 𝑀_s = 2.31 (4.3)

where 𝑆_𝑚𝑖𝑛(𝑠) expresses the low inertia system. Scaling to individual units is explained in Section 5.

Example:

To exemplify the robustness, study the margin for the increased regulating strength. Stability requires the point -1 to not be encircled. Therefore, consider the loop point that has 180^o phase shift (F(jω1)∙G(jω1)=-1) pointing in the negative direction along the real axis. Assume the point to lie just at the circle, i.e. the coordinate is 0j+(r-1). Then, the loop gain is written as a function of the regulating strength as follows

𝐺₀|_𝜔1= 𝐹(𝑗𝜔₁)𝐺(𝑗𝜔₁) =(𝑅₀+ ∆𝑅)

6000 (1 − 𝑟)𝑒^𝑗𝜋 ^(4.4)

where R0=6000 MW/Hz is the regulating strength used in the design and ∆R is the additional regulating strength.

This corresponds to scale the regulating strength as follows (6000 + ∆𝑅)

6000 (1 − 𝑟)𝑒^𝑗𝜋 > −1 (4.5)

Then, ∆𝑅 < 4582 MW/Hz

Note that the frequency will oscillate and the quality may be poor but stability is ensured.

Additional regulating strength coming from backlash is not included. Since the capacity is reduced from backlash the regulating strength in terms of MW per Hz will increase as backlash comes in to play. This is explained by the fact that the procured capacity is the steady state capacity.

P

ERFORMANCE REQUIREMENT

4.2

To develop performance requirements the disturbance needs to be quantified. The disturbance is here net-power variations in normal operation that are to be balanced by the FCR-N. This variation was estimated by accessing the energy metering system that Svenska kraftnät operates. Within this energy metering system, all transfers between the grid owned by Svenska kraftnät and a third party are monitored and logged with sufficient accuracy and with a sampling rate of three seconds. The system also includes energy meters for all tie-lines connecting between different

(24)

bidding areas within Sweden. The tie-lines used to measure the net imbalances of a larger area were the AC tie-lines interconnecting Areas SE3 and SE4, see Figure 18. This area was measured because there is a very small amount of FCR-N active within this area, giving the measured values a high degree of relevancy for the underlying stochastic generation-load imbalances. Also, the load within the southern Swedish area constitutes on average a third of the total load in the Nordic system. The data processing and detailed results are provided by the Imbalance study, see [5].

FIGURE 18.SCHEMATIC DESCRIPTION OF WHICH TIE-LINES MEASURED TO ESTIMATE GENERATION-LOAD IMBALANCES.MEASURED CUTS ARE THOSE SHOWN WITH A RED LINE IN THE MAP ON RIGHT HAND SIDE.

The study aimed to emulate the statistical properties of the measured net-variation by modelling it as the output of a linear filter with white noise as input. The study estimated the variations to have low-pass characteristics and the process is given by

𝑑 = 𝐷_est(𝑠)𝑤 =√3 ∙ 12

𝑠 𝑤 (4.6)

where d is the net-power variation and w is the white noise input to the filter Dest(s). The aFRR also contributes to balance the system and with its integration balancing from 2-3 minutes in addition to the tertiary frequency control, manual frequency restoration reserve. In steady-state the capacity is specified to 600 MW and therefore the imbalance profile is mapped to a first order filter as

(25)

𝐷(𝑠) = 600 𝑇_dist𝑠 + 1

(4.7) where Tdist is the time constant of the imbalance profile.

Measurement data were not available over longer periods to estimate net-power variations in order to verify the spread.

Eq. (2.7) states how a disturbance propagates through the system and is now written as 𝑆_avg(𝑠)𝐷(𝑠)𝐺_avg(𝑠)𝑤 = 𝑓

(4.8) where Gavg(s) is the transfer function of the average inertia system. Rewriting this as

𝑆_avg(𝑠)𝑤 =_{𝐷(𝑠)𝐺}¹

avg(𝑠)𝑓. (4.9)

The power spectral density (PSD) state the relation between the input signal and the output given as

|𝑆_avg(𝑗𝜔)|²∅_𝑤(𝑗𝜔) =_{|𝐷(𝑗𝜔)𝐺}¹

avg(𝑗𝜔)|²∅_f(𝑗𝜔) (4.10)

where ∅_w(𝑗𝜔) is white noise with PSD equal to one and ∅_f(𝑗𝜔) is the PSD of the frequency deviation.

An output with equal limitation at each frequency and the PSD is constant ∅_f(𝑗𝜔) = 𝜎_f² is required.

This choice specifies a boundary of the amplification of all frequencies and a variance of the frequency deviation. The performance requirement then becomes

|𝑆_𝑎𝑣𝑔(𝑠)| <_{|𝐷(𝑠)𝐺}^𝜎^f

avg(𝑠)|= 𝜎_𝑓_{𝐷(𝑗0)|𝐺}^|𝑇^dist^𝑠+1|

avg(𝑠)| . (4.11)

From this, the steady-state value is 6000 MW/Hz which is the ratio between 𝜎_𝑓 and D(j0). As mentioned before, deterministic disturbance signals require that (4.11) is fulfilled (with 𝜎_f= 0.1) in order to let an input signal of amplitude 600∙sin(ωt) MW (=1 pu) not result in a frequency deviation larger than 0.1∙sin(ωt) Hz (=1 pu).

Enforcing the frequency target to 0.1 Hz/Hz at all frequencies and select the time constant to align the transfer function in (4.11) does not necessarily ensure the frequency within ±0.1 Hz. As argued above, stochastic signals are better described by the statistical property. If the output frequency has the characteristics of white noise the frequency target have to be significantly reduced. Fortunately this is not the case as the bandwidth of the output is bounded since the inertia of the system reduces the effect on the output at high frequencies - what matter is the variance of the output frequency.. The power spectrum does not necessary have to be smaller than 0.1 Hz/Hz for all frequencies as it appears smaller at other frequencies.

In order to match the disturbance spectrum, and to still be able to obtain acceptable frequency quality, an appropriate time constant must be found. The filter constant is found through analysis

(26)

of a huge amount of simulation runs with various parameters sweeps that were performed in the Nordic frequency model.

In order to reduce the quadratic sum of the frequency deviation, which relates to variance of the output frequency, the approach above is not most appropriate. However, the approach specified above has advantage when it comes to real testing and implementation as it is straightforward to put requirement at particular time periods. Also, linear optimisation, see Appendix B, was performed to find parameters (Kp and Ki) for the linear reference unit that fulfilled the requirements. It was shown that there is a correlation between the resonance peak of the sensitivity function and frequency quality. This peak is directly related to the stability margin.

The imbalance study indicated that the disturbance could be mapped to a low pass filter. There is a trade-off between the filter constant and frequency quality while harder requirements result in less capacity on the market. The simulations are performed with the Nordic frequency model- profile by parameter sweeps over, Kp, Ti, backlash and droop. Then the minutes outside normal band, described in next subsection, are quantified.

R

EQUIREMENTS

4.3

There are several aspects to consider when deciding the filter constant of the disturbance filter. As described earlier, backlash has a great impact on the stability and performance. Thus, the signal strength plays an important role and there is a trade-off between this and the filter time constant.

Figure 19 illustrates an example of the sensitivity functions for specific parameters of the linear hydro power model.

FIGURE 19.ILLUSTRATION OF REQUIREMENTS AND PLOTTED SENSITIVITY FUNCTIONS

Note that, inspection of Figure 19 clearly shows typical margins between the sensitivity function of the low inertia system and the performance requirement. The low inertia system, instead, is limited by the stability requirement. Note that, the performance requirement is here plotted based on the average inertia system. The slope in the performance curve is moved to the right with decreased inertia.

To create a picture of the trade-off, a huge amount of simulations were performed. Figure 19 indicates that the performance requirement is close to be violated around ω=10^-2 rad/s and the

𝑺𝒎𝒊𝒏(𝒔) 𝑺𝒂𝒗𝒈(𝒔) Stability req.

Performance req.

(27)

stability requirement around ω=6∙10^-1 rad/s. Thus, performance is the limiting factor at longer time periods (≈600-200 s) and stability at shorter time periods (≈60-10 s).

In the beginning of the project 30 mHz was proposed for testing, with time it turned out backlash had too much impact compared to the units’ response.

Since the signal strength has great impact it has to be coordinated in order to find reasonable over-all requirements. It turned out that impact on the performance from backlash occurred at longer time periods where the phase lag of the FCR-response still was low. The backlash is more or less fixed as it comes from mechanical parts and is here specified in per unit, as described earlier.

From the tests performed with 50 mHz amplitude it was seen that it is only possible to fulfil the requirements for backlash up to ±0.004 pu, shown in Table 4. As set of parameters is a combination of Kp and Ki, the range of the parameters simulated in Table 4 is

Ki=[0.1, 0.15, 0.2, 0.25, 0.3, 0.4]

and

Kp=[1, 1.5, 2, 2.5, 3, 3.5, 4, 4.5, 5, 5.5, 6, 6.5, 7, 7.5, 8, 8.5, 9, 9.5, 10]

which leads to 72 combinations. In addition to this, a sweep is run over backlash and droop. In total 4752 number of qualification runs were performed.

(28)

TABLE 4.NUMBER OF QUALIFIED PARAMETER SETS (MAXIMUM 72) AS TIME CONSTANT OF THE DISTURBANCE FILTER VARIES BETWEEN 70 S – 100 S.

One can argue, if the backlash is ±10 % for an input amplitude of ±100 mHz, i.e. the maximum output is 90 %. Then if reducing the input amplitude to ±50 mHz, the maximum output becomes 80 %. Clearly, the loss in amplitude has increased by a factor of two. Supported by this argument, and the fact that only a low value of the backlash was allowed, an amplitude of ±100 mHz was chosen for performance^v requirements to reduce the impact from the backlash.

The impact from backlash on the stability requirement is more complex as both the amplitude and phase lag are reduced at the time periods of interest. A first attempt was to use amplitude of 50 mHz. This in order to capture instability in the range of small variation of the input which is the normal variation of today in the Nordic power system.

In order to decide a proper time constant for performance another round of simulations were performed on the Nordic frequency model. The results when varying backlash and droop are

v Also for stability – motivated by the fact that it will make the actual testing simpler without affecting the results too much.

70 s

80 s

90 s

100 s

(29)

shown in Table 7 and Table 8 for variation of the droop. These were then compared to each other together with the MoNB. The parameter sweeps are shown in Table 5 where the number of combinations are 19∙10∙12∙6=13680. The parameters that are sweeped are Kp, Ti, droop and backlash. These are sweeped for each choice of performance time constant i.e. 50-90 s. Table 6 shows the percentage of qualified units’ parameters and Figure 20 shows the duration curve with 600 MW FCR-N for different time constants. The x-axis indicates the percentage of all qualified units producing MoNB that is lower than or equal to a certain value (y-axis). Based on all simulations and studying the MoNB the time constant was selected to 70 s.

Note that Ti is here defined as 𝑇_𝑖 =_𝑒¹

p𝐾_i. (4.12)

The control structure used in the models has Ki implemented, see Figure 9. In the simulations Ki is scaled with ep so Ti becomes the same for any droop. The base case used is with a droop of 6 %.

TABLE 5.PARAMETER RANGES USED IN THE SIMULATION STUDY.

Parameter Step size Interval

Kp 0.5 1-10

Ti 10 s 10-100 s

Droop 2% 2-12%

Backlash 0.001 pu 0-0.012 pu

TABLE 6.SHARE OF COMBINATIONS THAT QUALIFY

Time constant Share that qualified

50 s 6.32%

60 s 9.81%

70 s 13.45%

80 s 17.18%

90 s 20.30%